1 /* Copyright (c) 2013 Coraid, Inc. See COPYING for GPL terms. */
2 /*
3 * aoeblk.c
4 * block device routines
5 */
6
7 #include <linux/kernel.h>
8 #include <linux/hdreg.h>
9 #include <linux/blk-mq.h>
10 #include <linux/backing-dev.h>
11 #include <linux/fs.h>
12 #include <linux/ioctl.h>
13 #include <linux/slab.h>
14 #include <linux/ratelimit.h>
15 #include <linux/genhd.h>
16 #include <linux/netdevice.h>
17 #include <linux/mutex.h>
18 #include <linux/export.h>
19 #include <linux/moduleparam.h>
20 #include <linux/debugfs.h>
21 #include <scsi/sg.h>
22 #include "aoe.h"
23
24 static DEFINE_MUTEX(aoeblk_mutex);
25 static struct kmem_cache *buf_pool_cache;
26 static struct dentry *aoe_debugfs_dir;
27
28 /* GPFS needs a larger value than the default. */
29 static int aoe_maxsectors;
30 module_param(aoe_maxsectors, int, 0644);
31 MODULE_PARM_DESC(aoe_maxsectors,
32 "When nonzero, set the maximum number of sectors per I/O request");
33
aoedisk_show_state(struct device * dev,struct device_attribute * attr,char * page)34 static ssize_t aoedisk_show_state(struct device *dev,
35 struct device_attribute *attr, char *page)
36 {
37 struct gendisk *disk = dev_to_disk(dev);
38 struct aoedev *d = disk->private_data;
39
40 return snprintf(page, PAGE_SIZE,
41 "%s%s\n",
42 (d->flags & DEVFL_UP) ? "up" : "down",
43 (d->flags & DEVFL_KICKME) ? ",kickme" :
44 (d->nopen && !(d->flags & DEVFL_UP)) ? ",closewait" : "");
45 /* I'd rather see nopen exported so we can ditch closewait */
46 }
aoedisk_show_mac(struct device * dev,struct device_attribute * attr,char * page)47 static ssize_t aoedisk_show_mac(struct device *dev,
48 struct device_attribute *attr, char *page)
49 {
50 struct gendisk *disk = dev_to_disk(dev);
51 struct aoedev *d = disk->private_data;
52 struct aoetgt *t = d->targets[0];
53
54 if (t == NULL)
55 return snprintf(page, PAGE_SIZE, "none\n");
56 return snprintf(page, PAGE_SIZE, "%pm\n", t->addr);
57 }
aoedisk_show_netif(struct device * dev,struct device_attribute * attr,char * page)58 static ssize_t aoedisk_show_netif(struct device *dev,
59 struct device_attribute *attr, char *page)
60 {
61 struct gendisk *disk = dev_to_disk(dev);
62 struct aoedev *d = disk->private_data;
63 struct net_device *nds[8], **nd, **nnd, **ne;
64 struct aoetgt **t, **te;
65 struct aoeif *ifp, *e;
66 char *p;
67
68 memset(nds, 0, sizeof nds);
69 nd = nds;
70 ne = nd + ARRAY_SIZE(nds);
71 t = d->targets;
72 te = t + d->ntargets;
73 for (; t < te && *t; t++) {
74 ifp = (*t)->ifs;
75 e = ifp + NAOEIFS;
76 for (; ifp < e && ifp->nd; ifp++) {
77 for (nnd = nds; nnd < nd; nnd++)
78 if (*nnd == ifp->nd)
79 break;
80 if (nnd == nd && nd != ne)
81 *nd++ = ifp->nd;
82 }
83 }
84
85 ne = nd;
86 nd = nds;
87 if (*nd == NULL)
88 return snprintf(page, PAGE_SIZE, "none\n");
89 for (p = page; nd < ne; nd++)
90 p += scnprintf(p, PAGE_SIZE - (p-page), "%s%s",
91 p == page ? "" : ",", (*nd)->name);
92 p += scnprintf(p, PAGE_SIZE - (p-page), "\n");
93 return p-page;
94 }
95 /* firmware version */
aoedisk_show_fwver(struct device * dev,struct device_attribute * attr,char * page)96 static ssize_t aoedisk_show_fwver(struct device *dev,
97 struct device_attribute *attr, char *page)
98 {
99 struct gendisk *disk = dev_to_disk(dev);
100 struct aoedev *d = disk->private_data;
101
102 return snprintf(page, PAGE_SIZE, "0x%04x\n", (unsigned int) d->fw_ver);
103 }
aoedisk_show_payload(struct device * dev,struct device_attribute * attr,char * page)104 static ssize_t aoedisk_show_payload(struct device *dev,
105 struct device_attribute *attr, char *page)
106 {
107 struct gendisk *disk = dev_to_disk(dev);
108 struct aoedev *d = disk->private_data;
109
110 return snprintf(page, PAGE_SIZE, "%lu\n", d->maxbcnt);
111 }
112
aoedisk_debugfs_show(struct seq_file * s,void * ignored)113 static int aoedisk_debugfs_show(struct seq_file *s, void *ignored)
114 {
115 struct aoedev *d;
116 struct aoetgt **t, **te;
117 struct aoeif *ifp, *ife;
118 unsigned long flags;
119 char c;
120
121 d = s->private;
122 seq_printf(s, "rttavg: %d rttdev: %d\n",
123 d->rttavg >> RTTSCALE,
124 d->rttdev >> RTTDSCALE);
125 seq_printf(s, "nskbpool: %d\n", skb_queue_len(&d->skbpool));
126 seq_printf(s, "kicked: %ld\n", d->kicked);
127 seq_printf(s, "maxbcnt: %ld\n", d->maxbcnt);
128 seq_printf(s, "ref: %ld\n", d->ref);
129
130 spin_lock_irqsave(&d->lock, flags);
131 t = d->targets;
132 te = t + d->ntargets;
133 for (; t < te && *t; t++) {
134 c = '\t';
135 seq_printf(s, "falloc: %ld\n", (*t)->falloc);
136 seq_printf(s, "ffree: %p\n",
137 list_empty(&(*t)->ffree) ? NULL : (*t)->ffree.next);
138 seq_printf(s, "%pm:%d:%d:%d\n", (*t)->addr, (*t)->nout,
139 (*t)->maxout, (*t)->nframes);
140 seq_printf(s, "\tssthresh:%d\n", (*t)->ssthresh);
141 seq_printf(s, "\ttaint:%d\n", (*t)->taint);
142 seq_printf(s, "\tr:%d\n", (*t)->rpkts);
143 seq_printf(s, "\tw:%d\n", (*t)->wpkts);
144 ifp = (*t)->ifs;
145 ife = ifp + ARRAY_SIZE((*t)->ifs);
146 for (; ifp->nd && ifp < ife; ifp++) {
147 seq_printf(s, "%c%s", c, ifp->nd->name);
148 c = ',';
149 }
150 seq_puts(s, "\n");
151 }
152 spin_unlock_irqrestore(&d->lock, flags);
153
154 return 0;
155 }
156
aoe_debugfs_open(struct inode * inode,struct file * file)157 static int aoe_debugfs_open(struct inode *inode, struct file *file)
158 {
159 return single_open(file, aoedisk_debugfs_show, inode->i_private);
160 }
161
162 static DEVICE_ATTR(state, 0444, aoedisk_show_state, NULL);
163 static DEVICE_ATTR(mac, 0444, aoedisk_show_mac, NULL);
164 static DEVICE_ATTR(netif, 0444, aoedisk_show_netif, NULL);
165 static struct device_attribute dev_attr_firmware_version = {
166 .attr = { .name = "firmware-version", .mode = 0444 },
167 .show = aoedisk_show_fwver,
168 };
169 static DEVICE_ATTR(payload, 0444, aoedisk_show_payload, NULL);
170
171 static struct attribute *aoe_attrs[] = {
172 &dev_attr_state.attr,
173 &dev_attr_mac.attr,
174 &dev_attr_netif.attr,
175 &dev_attr_firmware_version.attr,
176 &dev_attr_payload.attr,
177 NULL,
178 };
179
180 static const struct attribute_group aoe_attr_group = {
181 .attrs = aoe_attrs,
182 };
183
184 static const struct attribute_group *aoe_attr_groups[] = {
185 &aoe_attr_group,
186 NULL,
187 };
188
189 static const struct file_operations aoe_debugfs_fops = {
190 .open = aoe_debugfs_open,
191 .read = seq_read,
192 .llseek = seq_lseek,
193 .release = single_release,
194 };
195
196 static void
aoedisk_add_debugfs(struct aoedev * d)197 aoedisk_add_debugfs(struct aoedev *d)
198 {
199 char *p;
200
201 if (aoe_debugfs_dir == NULL)
202 return;
203 p = strchr(d->gd->disk_name, '/');
204 if (p == NULL)
205 p = d->gd->disk_name;
206 else
207 p++;
208 BUG_ON(*p == '\0');
209 d->debugfs = debugfs_create_file(p, 0444, aoe_debugfs_dir, d,
210 &aoe_debugfs_fops);
211 }
212 void
aoedisk_rm_debugfs(struct aoedev * d)213 aoedisk_rm_debugfs(struct aoedev *d)
214 {
215 debugfs_remove(d->debugfs);
216 d->debugfs = NULL;
217 }
218
219 static int
aoeblk_open(struct block_device * bdev,fmode_t mode)220 aoeblk_open(struct block_device *bdev, fmode_t mode)
221 {
222 struct aoedev *d = bdev->bd_disk->private_data;
223 ulong flags;
224
225 if (!virt_addr_valid(d)) {
226 pr_crit("aoe: invalid device pointer in %s\n",
227 __func__);
228 WARN_ON(1);
229 return -ENODEV;
230 }
231 if (!(d->flags & DEVFL_UP) || d->flags & DEVFL_TKILL)
232 return -ENODEV;
233
234 mutex_lock(&aoeblk_mutex);
235 spin_lock_irqsave(&d->lock, flags);
236 if (d->flags & DEVFL_UP && !(d->flags & DEVFL_TKILL)) {
237 d->nopen++;
238 spin_unlock_irqrestore(&d->lock, flags);
239 mutex_unlock(&aoeblk_mutex);
240 return 0;
241 }
242 spin_unlock_irqrestore(&d->lock, flags);
243 mutex_unlock(&aoeblk_mutex);
244 return -ENODEV;
245 }
246
247 static void
aoeblk_release(struct gendisk * disk,fmode_t mode)248 aoeblk_release(struct gendisk *disk, fmode_t mode)
249 {
250 struct aoedev *d = disk->private_data;
251 ulong flags;
252
253 spin_lock_irqsave(&d->lock, flags);
254
255 if (--d->nopen == 0) {
256 spin_unlock_irqrestore(&d->lock, flags);
257 aoecmd_cfg(d->aoemajor, d->aoeminor);
258 return;
259 }
260 spin_unlock_irqrestore(&d->lock, flags);
261 }
262
aoeblk_queue_rq(struct blk_mq_hw_ctx * hctx,const struct blk_mq_queue_data * bd)263 static blk_status_t aoeblk_queue_rq(struct blk_mq_hw_ctx *hctx,
264 const struct blk_mq_queue_data *bd)
265 {
266 struct aoedev *d = hctx->queue->queuedata;
267
268 spin_lock_irq(&d->lock);
269
270 if ((d->flags & DEVFL_UP) == 0) {
271 pr_info_ratelimited("aoe: device %ld.%d is not up\n",
272 d->aoemajor, d->aoeminor);
273 spin_unlock_irq(&d->lock);
274 blk_mq_start_request(bd->rq);
275 return BLK_STS_IOERR;
276 }
277
278 list_add_tail(&bd->rq->queuelist, &d->rq_list);
279 aoecmd_work(d);
280 spin_unlock_irq(&d->lock);
281 return BLK_STS_OK;
282 }
283
284 static int
aoeblk_getgeo(struct block_device * bdev,struct hd_geometry * geo)285 aoeblk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
286 {
287 struct aoedev *d = bdev->bd_disk->private_data;
288
289 if ((d->flags & DEVFL_UP) == 0) {
290 printk(KERN_ERR "aoe: disk not up\n");
291 return -ENODEV;
292 }
293
294 geo->cylinders = d->geo.cylinders;
295 geo->heads = d->geo.heads;
296 geo->sectors = d->geo.sectors;
297 return 0;
298 }
299
300 static int
aoeblk_ioctl(struct block_device * bdev,fmode_t mode,uint cmd,ulong arg)301 aoeblk_ioctl(struct block_device *bdev, fmode_t mode, uint cmd, ulong arg)
302 {
303 struct aoedev *d;
304
305 if (!arg)
306 return -EINVAL;
307
308 d = bdev->bd_disk->private_data;
309 if ((d->flags & DEVFL_UP) == 0) {
310 pr_err("aoe: disk not up\n");
311 return -ENODEV;
312 }
313
314 if (cmd == HDIO_GET_IDENTITY) {
315 if (!copy_to_user((void __user *) arg, &d->ident,
316 sizeof(d->ident)))
317 return 0;
318 return -EFAULT;
319 }
320
321 /* udev calls scsi_id, which uses SG_IO, resulting in noise */
322 if (cmd != SG_IO)
323 pr_info("aoe: unknown ioctl 0x%x\n", cmd);
324
325 return -ENOTTY;
326 }
327
328 static const struct block_device_operations aoe_bdops = {
329 .open = aoeblk_open,
330 .release = aoeblk_release,
331 .ioctl = aoeblk_ioctl,
332 .compat_ioctl = blkdev_compat_ptr_ioctl,
333 .getgeo = aoeblk_getgeo,
334 .owner = THIS_MODULE,
335 };
336
337 static const struct blk_mq_ops aoeblk_mq_ops = {
338 .queue_rq = aoeblk_queue_rq,
339 };
340
341 /* alloc_disk and add_disk can sleep */
342 void
aoeblk_gdalloc(void * vp)343 aoeblk_gdalloc(void *vp)
344 {
345 struct aoedev *d = vp;
346 struct gendisk *gd;
347 mempool_t *mp;
348 struct request_queue *q;
349 struct blk_mq_tag_set *set;
350 ulong flags;
351 int late = 0;
352 int err;
353
354 spin_lock_irqsave(&d->lock, flags);
355 if (d->flags & DEVFL_GDALLOC
356 && !(d->flags & DEVFL_TKILL)
357 && !(d->flags & DEVFL_GD_NOW))
358 d->flags |= DEVFL_GD_NOW;
359 else
360 late = 1;
361 spin_unlock_irqrestore(&d->lock, flags);
362 if (late)
363 return;
364
365 gd = alloc_disk(AOE_PARTITIONS);
366 if (gd == NULL) {
367 pr_err("aoe: cannot allocate disk structure for %ld.%d\n",
368 d->aoemajor, d->aoeminor);
369 goto err;
370 }
371
372 mp = mempool_create(MIN_BUFS, mempool_alloc_slab, mempool_free_slab,
373 buf_pool_cache);
374 if (mp == NULL) {
375 printk(KERN_ERR "aoe: cannot allocate bufpool for %ld.%d\n",
376 d->aoemajor, d->aoeminor);
377 goto err_disk;
378 }
379
380 set = &d->tag_set;
381 set->ops = &aoeblk_mq_ops;
382 set->cmd_size = sizeof(struct aoe_req);
383 set->nr_hw_queues = 1;
384 set->queue_depth = 128;
385 set->numa_node = NUMA_NO_NODE;
386 set->flags = BLK_MQ_F_SHOULD_MERGE;
387 err = blk_mq_alloc_tag_set(set);
388 if (err) {
389 pr_err("aoe: cannot allocate tag set for %ld.%d\n",
390 d->aoemajor, d->aoeminor);
391 goto err_mempool;
392 }
393
394 q = blk_mq_init_queue(set);
395 if (IS_ERR(q)) {
396 pr_err("aoe: cannot allocate block queue for %ld.%d\n",
397 d->aoemajor, d->aoeminor);
398 blk_mq_free_tag_set(set);
399 goto err_mempool;
400 }
401
402 spin_lock_irqsave(&d->lock, flags);
403 WARN_ON(!(d->flags & DEVFL_GD_NOW));
404 WARN_ON(!(d->flags & DEVFL_GDALLOC));
405 WARN_ON(d->flags & DEVFL_TKILL);
406 WARN_ON(d->gd);
407 WARN_ON(d->flags & DEVFL_UP);
408 blk_queue_max_hw_sectors(q, BLK_DEF_MAX_SECTORS);
409 blk_queue_io_opt(q, SZ_2M);
410 d->bufpool = mp;
411 d->blkq = gd->queue = q;
412 q->queuedata = d;
413 d->gd = gd;
414 if (aoe_maxsectors)
415 blk_queue_max_hw_sectors(q, aoe_maxsectors);
416 gd->major = AOE_MAJOR;
417 gd->first_minor = d->sysminor;
418 gd->fops = &aoe_bdops;
419 gd->private_data = d;
420 set_capacity(gd, d->ssize);
421 snprintf(gd->disk_name, sizeof gd->disk_name, "etherd/e%ld.%d",
422 d->aoemajor, d->aoeminor);
423
424 d->flags &= ~DEVFL_GDALLOC;
425 d->flags |= DEVFL_UP;
426
427 spin_unlock_irqrestore(&d->lock, flags);
428
429 device_add_disk(NULL, gd, aoe_attr_groups);
430 aoedisk_add_debugfs(d);
431
432 spin_lock_irqsave(&d->lock, flags);
433 WARN_ON(!(d->flags & DEVFL_GD_NOW));
434 d->flags &= ~DEVFL_GD_NOW;
435 spin_unlock_irqrestore(&d->lock, flags);
436 return;
437
438 err_mempool:
439 mempool_destroy(mp);
440 err_disk:
441 put_disk(gd);
442 err:
443 spin_lock_irqsave(&d->lock, flags);
444 d->flags &= ~DEVFL_GD_NOW;
445 schedule_work(&d->work);
446 spin_unlock_irqrestore(&d->lock, flags);
447 }
448
449 void
aoeblk_exit(void)450 aoeblk_exit(void)
451 {
452 debugfs_remove_recursive(aoe_debugfs_dir);
453 aoe_debugfs_dir = NULL;
454 kmem_cache_destroy(buf_pool_cache);
455 }
456
457 int __init
aoeblk_init(void)458 aoeblk_init(void)
459 {
460 buf_pool_cache = kmem_cache_create("aoe_bufs",
461 sizeof(struct buf),
462 0, 0, NULL);
463 if (buf_pool_cache == NULL)
464 return -ENOMEM;
465 aoe_debugfs_dir = debugfs_create_dir("aoe", NULL);
466 return 0;
467 }
468
469